LCEL: The Pipe Operator

The pipe operator (|) composes Runnables into chains. Lazy evaluation, type safety, and the full Runnable interface.

Quick Reference

→chain = step_a | step_b | step_c — reads left to right
→Every step is a Runnable with .invoke(), .stream(), .batch()
→Chains are lazy — | just wires steps, .invoke() executes them
→RunnablePassthrough and RunnableLambda let you inject custom logic
→LCEL replaces LLMChain, ConversationChain, and all legacy chain classes

LCEL in LangChain v1

LCEL is still supported but de-emphasized in v1

In LangChain v1, the primary abstraction is create_agent (for agents) and LangGraph (for workflows). LCEL/Runnables remain in langchain-core for composing model calls, prompt templates, and parsers — but you no longer need LCEL chains for most agent tasks. Use LCEL for simple model pipelines (prompt → model → parser). Use create_agent or LangGraph for anything involving tool calling, memory, or multi-step reasoning.

The | Operator

LCEL (LangChain Expression Language) lets you compose any steps with |, just like Unix pipes. Left output becomes right input. Each step in the chain is a Runnable, and the pipe operator wires them together into a sequence that executes in order.

Every Step is a Runnable

Method	Description	Returns
.invoke(input)	Single sync call	Output
.stream(input)	Token-by-token sync	Iterator[Chunk]
.batch([inputs])	Parallel processing	List[Output]
.ainvoke(input)	Single async call	Output
.astream(input)	Token-by-token async	AsyncIterator[Chunk]
.astream_events(input, version="v2")	All intermediate events	AsyncIterator[Event]

Sign in to read this article

This is a premium article. Sign in with your Google account to continue.